Dr. Xinzhe Wu (武鑫喆)


  •   Google Scholar
  •   LinkedIn
  •   Github

    • I am a scientific researcher at Juelich Supercomputing Centre. I received my Ph.D. of Computer Science from Maison de la Simulation and University of Lille, France.

    Research

    Algorithms

    • Itertive methods for linear systems and eigenvalue problems
    • Machine learning

    High Performance Computing

    • High Performance Computing software engineering
    • HPC applications
    • Parallel Programming

    Projects

    Current Projects

    • PRACE 6IP WP8 - Performance portable linear algebra

    Past Projects

    • MYX - MUST Correctness Checking for YML and XMP Programs

    Publications

    PhD Thesis

    Xinzhe Wu. Contribution to the Emergence of New Intelligent Parallel and Distributed Methods Using a Multi-level Programming Paradigm for Extreme Computing. PhD thsis, University of Lille, 2019.  

    Conference Publications

    Journal Publications

    Technical Reports

    . . In , , pp. , , vol. , no. , pp. , , . [pdf] [codes] [project]

    Presentations

    Invited talks

    Serge G. Petiton and Xinzhe Wu. The Unite and Conquer GMRES-LS/ERAM method to solve sequences of linear systems. In International Workshop on Eigenvalue Problems: Algorithms; Software and Applications, in Petascale Computing (EPASA2018), Tsukuba, Japan, 5th-6th March 2018.

    Talks at conferences

    Serge G. Petiton and Xinzhe Wu. An Asynchronous Distributed and Parallel Unite and Conquer Method to Solve Sequences of Non-Hermitian Linear systems. In MATHIAS 2018: Computational Science Engineering and Data Science by TOTAL, Serris, France, October 2018.
    Xinzhe Wu, Serge G. Petiton and Yutong Lu. A Parallel Generator of Non-Hermitian Matrices computed from Given Spectra. In 13th International Meeting on High Performance Computing for Computational Science (VECPAR 18), São Pedro, Brazil, 17th-19th September 2018.
    Xinzhe Wu and Serge G. Petiton. A Parallel Generator of Non-Hermitian Matrices computed from Given Spectra. In 10th International Workshop on Parallel Matrix Algorithms and Applications (PMAA18), Zurich, Switzerland, 27th-29th Juin 2018.
    Xinzhe Wu and Serge G. Petiton. A Scalable Generator of Non-Hermintian Test Matrices Computed from Given Spectra for Large-scale Systems. In 3rd Workshop on Parallel Programming Models - Productivity and Applications, Aachen, Germany, 15th March 2018.
    Xinzhe Wu and Serge G. Petiton. A Parallel Generator of Non-Hermitian Matrices computed from Given Spectra. In SIAM Conference on Parallel Processing for Scientifc Computing, Tokyo, Japan, 7th-10th March 2018.
    Xinzhe Wu and Serge G. Petiton. A Distributed and Parallel Asynchronous Unite and Conquer Method to Solve Large Scale Non-Hermitian Linear Systems. In International Conference on High Performance Computing in Asia-Pacific Region (HPC Asia 2018), Tokyo, Japan, 28th-31st Jan. 2018.
    Serge G. Petiton, Xinzhe Wu and Tao Chang. A Distributed and Parallel Asynchronous Unite and Conquer Method to Solve Large Scale Non-Hermitian Linear Systems. In International Conference on Preconditioning Techniques for Scientific and Industrial Applications (Preconditioning 2018), Vancouver, Canada, 31st Jul.-2nd Aug. 2018.

    Posters

    Xinzhe Wu and Serge G. Petiton. Large Non-Hermitian Matrix Generation with Given Spectra. In International Workshop on Eigenvalue Problems: Algorithms; Software and Applications, in Petascale Computing (EPASA2018), Tsukuba, Japan, 5th-6th March 2018.

    Software

    Open Source Softwares

    • SMG2S: Scalable Matrix Generator with Given Spectra    

      • SMG2S is a software which provides to generator the non-Hermitian and non-Symmetric Matrices with User-customized eigenvalues. SMG2S is implemented in parallel based on MPI (Message Passing Interface) and C++11 to support efficiently the generation of test matrices on distributed memory platforms. We propse SMG2S to generate large matrices with known spectra to benchmark the iterative methods on supercomputers, whose convergence has strong connection with the properties of spectra.

    Codes

    • UCGLE: Unite and Conquer GMRES/LS-ERAM method  

      • UCGLE is a distributed and parallel implementation of iterative method based on C/MPI and PETSc to solve very large non-Hermitian linear systems in large-scale homogenous and heterogeneous clusters. UCGLE is a special variant of hybrid iterative methods preconditioned by Least-Squares polynomial. It uses the static MPI communicator to support the asynchronous communication.

    • m-UCGLE: multiple Unite and Conquer GMRES/LS-Eigensolver method  

      • m-UCGLE is an extension of UCGLE based on C++/MPI, Trilinos/Belos and Trilinos/Anasazi to solve large non-Hermitian linear systems with multiple Right-hand Sides at the same time on large-scale homogenous and heterogeneous clusters. Multiple block GMRES and block Krylov-schur computing components can be allocated simultaneously and managed by a specified scheduler. Convergence of m-UCGLE is accelerated by a block version of Least Squares Polynomial preconditioner developed during my thesis.